Evidano is an AI-powered qualitative data analysis platform that helps teams turn transcripts, logs, and charts into rapid, operational insights. A new quasi-experimental mixed-methods study published June 29, 2026 shows a multilevel onsite training plus a 6-month mentorship program in Northwest Ethiopia (Jan–Sep 2024) raised provider knowledge to ≈91/100, tripled HEW referrals, and cut diagnostic interval by 54.3% (from 56.5 to 25.8 days), see the full paper at PLOS Medicine. For researchers and program teams, the core question is not only whether outcomes improved, but which implementation mechanisms (leadership, mentorship cadence, pictorial HEW materials) drove change. This post shows how to run an AI-enabled, reproducible qualitative analysis of training interventions and map findings to operational decisions using Evidano (Evidano). Ethics note: this analysis approach is research-focused and non-diagnostic, patient data must remain de-identified and consented.
Key Takeaways
Multilevel onsite training combined with a 6-month mentorship reduced facility diagnostic interval by 54.3% and raised provider knowledge to about 91/100 in the Ethiopia pilot.
- The program trained 1, 020 health extension workers (HEWs) and delivered immersive hospital and primary-care courses between Jan–Sep 2024.
- Median knowledge scores rose from 54.6 to 90.9 (primary) and from 36.4 to 90.9 (secondary); diagnostic interval fell from 56.5 to 25.8 days (−54.3%).
- Budget was USD 52, 762, and program adaptations (mentorship cadence, pictorial HEW modules) mattered amid conflict and supply shortages.
Findings snapshot
| Metric | Value | Source / Date | Implication |
|---|---|---|---|
| HEWs trained | 1, 020 | Study (Jan–Mar 2024) | Large community reach; pictorial materials enabled low-literacy teaching |
| Knowledge (median), primary & secondary | From 54.6 → 90.9 (primary); 36.4 → 90.9 (secondary) | KAP pre/post (6 months) | Substantial cognitive gain across tiers |
| Diagnostic interval (median) | 56.5 days → 25.8 days (−54.3%) | Chart review of 100 patients | Major reduction in facility diagnostic delay |
| Patient delay (symptom → first contact) | 27.0 days → 24.5 days (−9.3%) | Chart review | Smaller change, needs community awareness scaling |
| Budget | USD 52, 762 | Study report | Relatively low-cost with measurable system effects |
Fast take + source
A University of Gondar-led program (Jan–Sep 2024) delivered immersive onsite courses for clinicians, pictorial modules for 1, 020 health extension workers, and a 6-month mentorship program, producing large knowledge and practice gains and a 54.3% reduction in diagnostic interval.
- Design: quasi-experimental pre–post mixed-methods with KAP surveys, chart reviews (n=100 patient charts), in-depth interviews (IDIs) and focus group discussions (FGDs).
- Operational constraints: active regional conflict (April 2023 onward), seasonal rains, and reagent shortages were tracked in implementation logs.
What happened (plain English)
The intervention combined immersive 10-day training for hospital clinicians, seven-day workshops for primary providers, and a five-day pictorial curriculum for HEWs, followed by monthly (then biweekly) mentorship and tele-support through April–September 2024.
- Process fidelity: 90% of planned training days were delivered and 87% attended all sessions.
- Quantitative impact: large KAP increases and tripled HEW referrals (0.4 → 1.2 referrals/HEW/month).
- Context: supply shortages and strained tertiary capacity limited speed-to-treatment despite faster diagnosis.
Qualitative analysis of training interventions: what to code
Mechanisms to prioritize
Prioritize coding mentorship cadence (monthly vs biweekly), leadership engagement, referral-form use, and HEW community activities because these map directly to maintenance of knowledge and referral quality.
Code quotes about barriers (reagents, transport, cultural beliefs) that explain why diagnostic gains did not immediately translate into faster treatment initiation.
Sampling & triangulation
Use purposive IDIs with mentors, facility managers, and HEWs plus FGDs to capture social norms, and triangulate findings with KAP score shifts and chart-derived intervals to strengthen causal claims in a pre–post design.
Tag timeline references in transcripts (for example, June 2024 mini-trainings) to link adaptations to outcome inflections.
So what for researchers and managers
Implementation researchers
Implementation researchers should use mixed-methods coding to test which CFIR domains (inner setting, process, characteristics of individuals) correlate with sites retaining higher knowledge (95.2% vs 88.3%).
Prioritize longitudinal thematic analysis to detect attitude fatigue, such as the modest decline in attitude scores, and design remedial motivational modules.
Program managers / MOH
Program managers and ministries of health should scale the pictorial HEW modules and formalize cascade trainings for conflict-affected areas, and monitor referral-form error rates (reduced 60% vs 35% with extra support) as a leading indicator.
Invest in diagnostic supplies in parallel, because diagnostic gains without treatment capacity will bottleneck outcomes.
Analysts & evaluators
Analysts and evaluators should treat pre–post designs with qualitative anchors, using a codebook-driven thematic analysis to explain plausible mechanisms when controls are absent.
Flag and code external campaigns (for example, malaria control) and security events to adjust attribution.
Do more, faster with Evidano
Ingest and prepare all materials
Upload transcripts, training slides, implementation logs, and HEW pictorial feedback into Evidano to prepare a searchable corpus for analysis.
Use auto-transcription with a custom dictionary to preserve local terms (for example, “nififit”) and enable PII redaction to maintain research ethics.
Automated thematic + frequency analysis
Run thematic extraction across IDIs and FGDs and link theme frequencies to site-level KAP changes; Evidano returns code→subcode hierarchies and co-occurrence networks so you can see which barriers cluster with diagnostic delays.
Identify the handful of high-leverage mechanisms, such as mentorship frequency and reagent supply, in minutes rather than weeks.
Cross-segment & timeline analyses
Compare themes across facility tiers, districts with or without cascade trainings, and pre/post adaptations (for example, June 2024 mini-trainings) to see which changes coincide with retained knowledge (95.2% vs 88.3%).
Use cross-segment analyses for faster hypothesis testing to support scale decisions.
Shareable visuals & quotes for stakeholders
Export co-occurrence networks, hierarchical code maps, and vetted, de-identified quotes to stakeholder briefs or MOH presentations to support evidence-forward storytelling.
Use these outputs to ease buy-in for diagnostic supplies and budget requests.
Security and compliance
Evidano encrypts data end-to-end and supports PII redaction, and customer data is not used to train third-party models, enabling secure program evaluations in fragile settings.
Maintain a custom dictionary to preserve local terms while protecting participant privacy.
Checklist: 7-step workflow to reproduce this analysis in Evidano
This seven-step checklist reproduces the study analysis using transcripts, charts, and logs. Step 1: Gather inputs, including KAP surveys, 100 patient charts, IDI/FGD audio, mentorship logs, and training agendas (Jan–Sep 2024). Step 2: Upload and preprocess, use Evidano transcription with custom dictionary entries and enable PII redaction. Step 3: Define a codebook, import CFIR-based codes or start with auto-suggested themes and refine. Step 4: Run thematic and frequency analysis, generate theme counts, co-occurrence matrices, and timeline-coded events (for example, June 2024 adaptations). Step 5: Run cross-segment analysis, comparing themes by facility tier, district, and mentorship cadence. Step 6: Validate, pulling representative de-identified quotes, checking intercoder consistency, and reconciling qualitative findings with KAP and chart outcomes. Step 7: Deliver, export visualizations and an executive brief for MOH with clear operational recommendations (for example, increase tele-support cadence and procure reagents).
FAQ: qualitative analysis of training interventions
How do I compare segments reliably?
Compare segments reliably by using balanced sampling and cross-segment comparison tools to normalize for transcript length and participant counts, and always triangulate with quantitative markers such as KAP scores and referrals.
Normalize for differences in participation and use timeline anchors when interpreting differences across districts and facility tiers.
Can AI handle local language terms?
Yes, preserve local language terms by using a custom dictionary so translations keep clinical meaning, and verify key term mappings against native speakers in the team.
Tag local terms for consistent handling in coding and quote extraction.
How secure is the platform for sensitive health data?
Evidano encrypts data, supports PII redaction, and does not use customer data to train third-party models, making it suitable for program evaluations in fragile settings.
Apply role-based access and export controls when sharing de-identified quotes with stakeholders.
Wrapping up & next steps
The PLOS Medicine study (published June 29, 2026) demonstrates that multilevel onsite training plus mentorship can cut diagnostic delays substantially, and rapid explanatory qualitative analysis is required to translate training into system fixes such as supply chain improvements, motivation interventions, and referral coordination.
- If you are evaluating trainings or scaling mentorship, use AI-enabled qualitative analysis to connect participant experiences to measurable outcomes quickly.
- Ready to reproduce these analyses on your corpus? Try Evidano for free to upload your transcripts, logs, and charts and get a thematic and cross-segment report in hours, not weeks.